1 / 48

Learning: Overview and Decision Trees

Learning: Overview and Decision Trees. Learning: Intro and General concepts Decision Trees Basic Concepts Finding the best DT. Agents in the World. Problem Solving/Planning Agent Architecture. Problem/Task. Reasoning Procedure. Internal representation. Background Knowledge/. Answer/

elia
Télécharger la présentation

Learning: Overview and Decision Trees

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Learning: Overview and Decision Trees • Learning: Intro and General concepts • Decision Trees • Basic Concepts • Finding the best DT

  2. Agents in the World

  3. Problem Solving/Planning AgentArchitecture Problem/Task Reasoning Procedure Internal representation Background Knowledge/ Answer/ Performance

  4. Learning • Learning is the ability to improve one’s behavior • based on experience. • The range of behaviors is expanded: the agent can do more. • The accuracy on tasks is improved: the agent can do things better. • The speed is improved: the agent can do things faster.

  5. Components of the Learning Problem • Task: the behavior or task that’s being improved. • For example: classification, acting in an environment • Data: the experiences that are being used to improve performance in the task. • Measure of improvement: How can the improvement be measured? • For example: increasing accuracy in prediction, new skills that were not present initially, improved speed.

  6. Learning AgentArchitecture Problem/Task Experience/Data Internal representation Reasoning Procedure Induction Procedure Background Knowledge/ Bias Answer/ Performance

  7. Learning as an Induction Problem • Induction: inferring a representation given a set of examples • EX: {q(a) , p(a)} • {q(b), p(b)} q(x) ← p(x) or • {q(c), p(c)} p(x) ← q(x) or neither? • what if I also observe {q(d)}?

  8. Common Learning Tasks • Supervised classification: given a set of pre-classified training examples, classify a new instance. • Unsupervised classification: find natural classes for examples. • Reinforcement learning: determine what to do based on rewards and punishments.

  9. Common Learning Tasks (1) • Supervised classification: given a set of pre-classified training examples, classify a new instance. • Alan does not drink coffee with sugar • Alan does not drink tea with sugar • Alan drinks diet lemonade • Will Alan drink a diet coke?

  10. Example Classification Data (2): reading newsgroup postings We want to classify new examples on property Actionbased on the examples’ attributes Author, Thread, Length, and Where.

  11. Common Learning Tasks (2) • Unsupervised classification: find natural classes for examples. • e.g. given patterns of users selections of web pages find categorizations of “user types” • e.g. given patterns of interface actions in an Tutoring Systems, (e.g., lots of hints requested, lots of repeated errors…) predict success in learning • Reinforcement learning: determine what to do based on rewards and punishments. • e.g. A butler robot brings Alan coffee and the newspaper and gets a reward. Is it for the coffee or for the newspaper or both?

  12. Feedback • Learning tasks can be characterized by the feedback given • to the learner. • Supervised learning: what has to be learned is specified for each example. • Unsupervised learning: no classifications are given; the learner has to discover categories and regularities in the data. • Reinforcement learning: feedback occurs after a sequence of actions. • +

  13. Representation of Learned Information Problem/Task Experience/Data Internal representation Reasoning Procedure Induction Procedure Background Knowledge/ Bias Answer/ Performance

  14. Representation of Learned Information • Any of the representation schemes studied in 322 and here can be used to represent the learned info. • Logic-based propositions • Probabilistic schemes • Utility functions • ……. • .

  15. Learning: Overview and Decision Trees • Learning: Intro and General concepts • Supervised Learning • Decision Trees • Basic Concepts • Finding the best DT

  16. Supervised Learning • Supervised Learning => pure inductive inference • Given a set of examples of a function f: pairs (x, f(x)) • Find a function h that approximates f. • h is calledhypothesis.

  17. Supervised Learning • H is the space of hypotheses (representations) that we consider, e.g. • Linear functions only • polynomial functions • Sinusoidal functions (ax+b+c*sin(x)) • How does one chose?

  18. Supervised Learning • H is the space of hypotheses (representations) that we consider, e.g. • Linear functions only • polynomial functions • Sinusoidal functions (ax+b+c*sin(x)) • How does one chose? • Aim to reduce complexity: simpler representations are easier to learn and use • Aim to get enough expressiveness to learn the true function • E.g. Cannot learn from sinusoidal data with polynomial functions

  19. Supervised Learning • The learning problem is called • Regression if f(x) is number • Classification, if f(x) is one of a finite set of values • Alan does not drink coffee with sugar • Alan does not drink tea with sugar • Alan drinks diet lemonade • Will Alan drink a diet coke?

  20. Measuring Success • The measure of success is not how well the agent performs on the training examples, but how well the agent performs for new examples. • Consider two agents: • Pclaims the negative examples seen are the only negative examples. Every other instance is positive. • Nclaims the positive examples seen are the only positive examples. Every other instance is negative. • Both agents correctly classify every training example, but disagree on every other example.

  21. Bias • Should we chose N’s or P’s hypothesis? • The data does not help us with the decision • The tendency to prefer one hypothesis over another is called a bias. • To have any inductive process make predictions on unseen data, you need a bias. • What constitutes a good bias is an empirical question about which biases work best in practice.

  22. Learning as Search • Given a representation and a bias, the problem of learning can be reduced to one of search. • Learning is search through the space of possible representations looking for the representation or representations that best fits the data, given the bias. • These search spaces are typically prohibitively large for systematic search. Use greedy heuristics. • A learning algorithm is made of a search space, an evaluation function, and a search method.

  23. Noise in the Data • Data isn’t perfect: • some of the attributes are assigned the wrong value • the attributes given are inadequate to predict the classification data • E.g., in Alan’s beverage preferences example adding whether it was sunny or cloudy when Alan got the drink • there are examples with missing attributes • e.g. in the beverage example, suppose that what matters to Alan is weather the drink is organic or not, but we don’t have this information for the given examples

  24. Noise in the Data • Overfitting occurs when there are random correlations in the training set, that do not appear in the unseen examples. • Alan does not drink HOT coffee with sugar • Alan does not drink HOT tea with sugar • Alan drinks COLD diet lemonade • Will Alan drink a COLD diet coke? • The system may say “YES” because “hot” appears in both negative examples and • “cold” appears in the positive one, but this is only do to chance. The correct • answer is “NO” because Alan does not like caffeinated beverages

  25. Assessing Performance of Learning Algorithm • Collect a large set of examples • Divide into two sets: training set and test set • Use the learning algorithm on the training set to generate an hypothesish • Measure the percentage of examples in the test set that are correctly classified by h • Repeat with training sets of different size and different randomly selected training sets.

  26. Cross-Validation • We want a large training set, to get more reliable models. • But this implies having a small test set • Not ideal, as it may give good or bad results just by luck • One common solution: k-fold cross validation • Partition the training set into k sets • Run the algorithm k times, each time (fold) using one of the k sets as the test test, and the rest as training set • Report algorithm performance as the average performance (e.g. ac curacy) over the k different folds • Useful to select different candidate algorithms/models • Once the algorithm/model type is selected via cross-validation, return the model trained on all available data

  27. Learning: Overview and Decision Trees • Learning: Intro and General concepts • Supervised Learning • Decision Trees • Basic Concepts • Finding the best DT

  28. Learning Decision Trees • Method for supervised classification (we will assume • attributes with finite discrete values) • Representation is a decision tree. • Bias is towards simple decision trees. • Search through the space of decision trees, from simple decision trees to more complex ones.

  29. Example Classification Data (2) We want to classify new examples on property Actionbased on the examples’ Author, Thread, Length, and Where.

  30. Learning task • Inductive inference • Given a set of examples of • f(author,thread, length, where) = {reads,skips} • Find a function h(author,thread, length, where) that approximates f

  31. Decision Trees (DT) • A decision tree is a tree where: • The non-leaf nodes are labeled with attributes. • The arcs out of a node labeled with attribute A are labeled with each of the possible values of the attribute A. • The leaves of the tree are labeled with classifications.

  32. new short long short Example Decision Tree (1) author unknown known length thread old reads skips length skips long skips reads

  33. DT as classifiers • To classify an example, filter in down the tree • For each attribute of the example, follow the branch corresponding to that attribute’s value. • When a leaf is reached, the example is classified as the label for that leaf.

  34. new short long short DT as classifiers author unknown known length thread old reads skips e4 e6 e2 e3 length skips long skips e1 reads e5

  35. Equivalent Rule Based Representation If author is unknown and length is short thenuser-action is reads If author is unknown and length is long thenuser-action is skips If author is known and thread is new and length is short thenuser-action is reads If author is known and thread is new and length is long thenuser-action is skips If author is known and thread is old thenuser-action is skips Suppose this is the true criteria that my user is employing

  36. Trivially Correct Decision Tree next author thread length l where

  37. Trivially Correct Decision Tree author PROBLEM? unknown known thread thread new old new old length length length length long short long short long where where where where where work home work work home home skips reads skips reads skips skips e2 e3 e1 e5 e6 e4

  38. Example Decision Tree (2) length

  39. long short Example Decision Tree (2) length e1, e3, e4, e6 e2, e5 read skips But this tree also classifies my examples correctly. Why? What do I need to get a tree that is closer to my true criterion

  40. Searching for a Good Decision Tree • Given some data, which decision tree should be generated? Search space is huge (more on this later) • Need a bias to guide the selection of a good tree. • Example, prefer the smallest tree. • Least depth? Fewest nodes? Which trees are the best predictors of unseen data? • How should one go about building a decision tree? The space of decision trees is too big for systematic search for the smallest decision tree.

  41. Searching for a Good Decision Tree • The input is • a target attribute for which we want to build a classification, • a set of examples • a set of attributes. • Stop if all examples have the same classification (good ending). • Plus some other stopping conditions for not so good endings • Otherwise, choose an attribute to split on (greedy, or myopic step) • for each value of this attribute, build a sub-tree for those examples with this attribute value

  42. known unknown new new old short long Building a Decision Tree (Newsgroup Read Domain) e1, e2, e3, e5, e4, e6 author, thread, length, where author e1, e5, e4, e6 thread, length, where e2, e3 thread thread old e4, e6 e1, e5 length, where length reads skips skips e2 e3 skips reads e5 e1

  43. Decision Tree Learning

  44. Decision Tree Learning

  45. Decision Tree Learning

  46. Decision Tree Learning

  47. known unknown new new old short long Building a Decision Tree (Newsgroup Read Domain) e1, e2, e3, e5, e4, e6 author, thread, length, where author e1, e5, e4, e6 thread, length, where e2, e3 thread thread old e4, e6 e1, e5 length, where length reads skips skips e2 e3 skips reads e5 e1

  48. Decision Tree Learning Plurality Value selects the most common output among a set of examples, breaking ties randomly

More Related